GenAI Daily - November 4, 2026: Snowflake's Always-On Agents, Apple Opens Siri to Developers, Anthropic's RSI Warning
Top Stories
Snowflake Adds Always-On Bots to CoCo and CoWork
At Expedition 2026 on November 3, Snowflake announced CoCo and CoWork Bots, always-on AI coworkers that work in the background, act in the tools teams already use, and check in only when a decision or sign-off is required.
Bots act on behalf of the user and inherit that user's access rights, role context, and permissions.
The Bots are coming soon to private preview. Other pieces ship sooner.
CoCo Automations is generally available and lets data teams schedule repeated workflows.
Every run keeps an audit trail and task history, so teams can suspend or rerun a job.
The surrounding releases target day-to-day use.
Snowflake for Slack, CoCo for Excel and Sheets, and Snowflake Mobile are in public preview.
An asynchronous API for CoWork lets longer, multi-step tasks continue in the background and resume if interrupted.
Snowflake said CoWork is used by more than 5,800 accounts and CoCo by more than 9,100.
The background-agent pattern matches ServiceNow's Arc, which can identify unfinished work and continue assignments for hours or days. Snowflake's difference is that the agents sit directly on governed warehouse data and inherit existing role permissions.
Snowflake press release | Unite.AI
Why it matters: Data teams can start scheduling audited recurring work now, and permission inheritance gives security reviewers a familiar model for evaluating the Bots once they reach preview.
Apple Opens Siri AI to Third-Party Apps in iOS 27.3 Beta
On November 3, Apple announced a Developer Preview in the iOS 27.3, iPadOS 27.3, macOS 27.3 and visionOS 27.3 betas that lets any app work with Siri AI.
Developers use App Intents to expose actions, and new skills chain several actions into one request.
The app must expose its own actions. The preview does not give Siri automatic access to everything in it.
Xcode's Siri Integration Advisor provides implementation feedback, and a Siri Evaluations framework helps developers test requests and track changes in responses.
The limits are important for planning.
Integrations can run in development-signed builds and versions shared with internal TestFlight testers, but not through external TestFlight testing or App Store releases.
Apple says the capabilities will first reach customers in a few months, with App Store submissions opening in spring 2027.
Apple is holding online labs for developers on November 4 and 5.
Why it matters: Teams with iOS apps have until spring to decide which actions to expose and to test them with the new evaluation framework before Siri becomes an entry point for their customers.
Anthropic S-1 Update: RSI Language and 90% AI-Written Code - UPDATE
Anthropic's amended S-1 adds new risk-factor language on recursive self-improvement.
The filing says Anthropic has already used AI to develop new models whose leaps in capability it didn't recognize at the time, and that it may soon meet its own definition of RSI.
Anthropic also said none of its models have crossed the threshold and that it will keep humans in the R&D process for the foreseeable future.
Internal metrics in the filing show Claude models authored more than 90% of the code merged into Anthropic's codebase.
Coverage is split on how to read it.
TechCrunch notes that an S-1 serves as both disclosure and marketing, so claims of being close to a self-building model could fall into the latter category.
Under Anthropic's internal safety frameworks, hitting the threshold triggers mandatory evaluations and potential revisions. This is separate from the financial figures covered earlier this week.
Why it matters: Buyers of Claude models should expect the Responsible Scaling Policy evaluations tied to this threshold to affect release timing and access controls on future models.
Key Developments
Nvidia Tells Cloud Partners to Keep Non-Chinese Revenue Above 51%
The Financial Times reported that Nvidia requires cloud providers to earn at least 51% of service revenue from non-Chinese customers.
The requirements cover new contracts worldwide, including data centers outside Asia, but exclude existing leasing agreements. The policy was shared verbally, and the report does not explain how the 51% is calculated or how Nvidia would audit it.
For neoclouds, the squeeze is practical.
A Singapore-based group said it is difficult to find enough non-Chinese clients outside the US, and that strict implementation would hurt business.
Nvidia neither confirms nor denies the policy.
Impact: Teams renting GPU capacity from smaller regional clouds should confirm their provider's customer mix and contract terms, since capacity expansion could be rationed.
DeepSeek Plans V5 Pre-Training on Huawei Ascend Hardware - UPDATE
Reports say DeepSeek intends to deploy at least 160,000 Huawei Ascend 950DT chips at a new Inner Mongolia site.
That fleet was initially meant for inference.
DeepSeek reportedly aims to start pre-training on the Ascend 950DT as soon as late 2026. The timing is tight.
Huawei is targeting commercial release of the chip in Q4 2026.
DeepSeek has reportedly struggled to train models on prior-generation Ascend chips because of stability and performance issues.
Software work is already underway.
On September 30, DeepSeek open-sourced infrastructure components adapted for Huawei's Ascend platform, including compiler tools, HPC libraries and distributed communication software. These are reports, not confirmed training results.
Crypto Briefing | Geopolitechs
Impact: If V5 trains cleanly on Ascend, open-weight Chinese models gain a supply path that export controls and the Nvidia cloud rule above cannot reach.
Canva Cuts 1,271 Roles as AI Serving Costs and Competition Bite
Canva announced a restructure affecting 1,271 people, or 22 percent of its headcount, through cuts, redefined roles or relocations.
The cost side is the notable part for builders.
Canva now expects revenue to grow about 20 percent in 2026, instead of the 30 percent it had planned.
The company had to slow the rollout of its big AI update, Canva AI 2.0.
Competitors include Lovable, which builds websites and apps from prompts, and AI labs such as Anthropic going after design with tools like Claude Design.
Trending Topics | Forbes Australia
Impact: Freemium products that ship AI features need per-task inference cost as a product constraint from the start, not an afterthought.
AMD Q3: Data Center Up 97% as MI450 Ramps
AMD reported third quarter revenue of $13.4 billion.
Data Center segment revenue was a record $8.5 billion, up 97% year-over-year, driven by EPYC processors and the launch of Instinct MI450 Series GPUs.
Management said it is guiding data center revenue up 50% sequentially in Q4, the quarter data center AI surpasses server business.
Q4 revenue guidance is about $17 billion, versus a consensus of roughly $16.38 billion.
Impact: A second GPU vendor shipping at volume gives teams more leverage when negotiating inference capacity, and Helios rack availability is worth tracking in Q4 planning.
Product Launches
Honeycomb AI SRE Agents
Honeycomb announced AI SRE agents that watch deploys, catch regressions before customers do, and pay down reliability debt.
Personal and Team Agents are in early access. Teams already using Honeycomb's MCP server for coding agents get a proactive layer on top of the same trace data.
Why it matters: Reliability work often gets deferred, and agents that monitor deploys and flag regressions early could help teams catch problems before customers notice, using trace data they already collect.
Databricks Genie Agents File Analysis
As of November 2, a Genie Agent can analyze unstructured files in Unity Catalog volumes, such as PDFs, slide decks, and images, and answer with citations pointing to the source page.
It is in public preview, and workspace admins control access from the Previews page. Source-page citations make answers easier to audit for document-heavy workloads.
Why it matters: Teams with large stores of PDFs, decks, and images can query them in place, and admin-controlled access plus page-level citations make it easier to govern and verify the answers.
Funding & Deals
Casap Raises $56M Series B
New York-based Casap sells AI agents to banks that investigate cases, make decisions and act across the systems institutions already use, applying each institution's own policies and bringing in people when a decision needs judgment.
The round brings total funding to $90 million, and revenue has grown tenfold since its $25 million Series A just over a year ago. It started with disputes and plans to expand into fraud, collections and claims. Led by Base10 Partners, with Experian and Synchrony participating.
Why it matters: Backing from Experian and Synchrony alongside tenfold revenue growth signals strong demand for agentic AI in regulated financial services workflows.
Salience Labs Raises $125M Series B
UK-based Salience Labs develops silicon photonics and optical circuit switching technology to speed up data center connectivity for AI workloads.
It will use the round to accelerate deployment, scale volume manufacturing and expand its U.S. presence in Silicon Valley. Investors include AMD Ventures and Intel Capital.
Why it matters: Interconnect is a growing bottleneck in AI data centers, and investment from AMD Ventures and Intel Capital shows chipmakers are backing optical approaches to ease it.
Drax, Brookfield and Radiant Plan Up to £20B Yorkshire AI Data Center
Investment in the full North Yorkshire development could exceed £20bn, and the initial phase of up to 100MW could receive £2.5bn in backing.
The Brookfield-Qai AI infrastructure joint venture will provide backing. The £20B figure is a ceiling. Radiant is seeking anchor customer commitments before starting construction.
Why it matters: The project shows how large UK AI capacity plans are, but with anchor customers still being sought, the £20B ceiling should be read as a potential rather than committed investment.
Tomorrow's Watch List
- Apple's online developer labs for the Siri AI preview continue November 5.
- OpenAI plans a Washington policy campaign timed to the lame-duck session beginning Nov. 9, centered on mandatory "speed limits" for the most powerful AI systems. Whether it becomes bill text is the thing to watch.
- Google's school AI and Search policy: Google has said search can be turned back on with AI Overviews after the backlash from lawmakers. Watch for further clarification of the admin controls.
- Snowflake reports fiscal Q3 2027 results on December 2, 2026.
Related reading: Check out this week's [Deep Insights analysis] for strategic context on these developments.
GenAI Daily is curated by rAInvent. For in-depth analysis, see the rAInvent articles.